Do not study the entire kit. Load it, answer the short interview, make one real asset, and judge the result. The product proves itself by producing work, not by making you admire seven markdown files.
Progress stays in this browser. Analytics records milestone names and blockers, not your Brand Profile, prompts, images, email, or business details.
1
Target: 3 minutes
Load the kit and make it activate.
Use a fresh ChatGPT or Gemini conversation. The reliable route is to unzip the download and upload every .md file together. Dropping the ZIP directly also works on models that unpack archives correctly.
Read every .md file inside in full - they are your instructions, not data to summarize. Follow them exactly, starting with the AI section of START-HERE.md.
If it replies with a summary instead of the interview, send this:
Greet me and start the interview.
Done when: the assistant greets you and begins asking about the business or project.
2
Target: 7 minutes
Answer the interview and save the Brand Profile.
Talk normally. The questions turn your positioning, audience, visual taste, practical constraints, and production needs into reusable decisions.
Give a concrete business or project, not a hypothetical category.
Share two or three references when visual taste is hard to describe.
Say what must remain exact: product shape, food, color, labels, people, price, or legal text.
When the Brand Profile appears, save it as a plain text or markdown file.
Done when: you have a saved Brand Profile file. That file is the memory bridge between sessions.
3
Target: 3 minutes
Choose one asset with one job.
Do not request a campaign, catalog, brand overhaul, and twelve-channel launch on the first run. Pick the smallest output that would be genuinely useful today.
Make one [channel] visual for [specific product, service, event, or idea]. Its job is [the business outcome]. Use my Brand Profile. Choose the safest production path, and tell me if you need a real reference photo before generating.
Good first assets include one Instagram product image, one menu special visual, one landing-page hero, one event poster substrate, or one concept poster. Small enough to finish, real enough to matter.
Done when: the request names one subject, one channel, and one business job.
4
Target: 8 minutes
Generate through the correct production path.
The first output should test the system, not your willingness to forgive the wrong workflow.
Exact subject
Start from a real photo
For an exact product, dish, room, person, label, or material, upload a real reference and ask Visual Engine to enhance or art-direct it.
Text-heavy asset
Separate image from exact type
Generate the visual substrate with safe text space. Add prices, dates, phone numbers, and dense copy in Canva or another editor.
Concept or mood
Generate directly
For campaign concepts, editorial scenes, or statement posters, let the system generate from the Brand Profile and chosen world.
Done when: one visual exists. It does not need to be perfect; it needs to be concrete enough to evaluate.
5
Target: 4 minutes
Verify first value, not mere completion.
This is the activation event. A finished interview is setup. A generated image is activity. First value means the output is useful enough to move real work forward.
Do not check a box to be polite. If one fails, use the blocker routes below, revise once, and judge again.
First value reached
You now have proof, not intention.
The next move is repetition around the same Brand Profile, not restarting the brand discussion every time.
Same asset type
Make the second one faster
Reuse the same world, camera, palette, and composition rules. Change only the subject, offer, or campaign variable that must change.
Choose the next engine
Expand only when the job demands it
Use Catalog Machine for batches, Promo Engine for offers, Concept Poster Engine for statements, and Brand Kit Builder when the whole system needs formalization.
Open the blocker that matches what happened. Each route changes the next action instead of sending another decorative reminder.
The assistant summarized the ZIP
It treated the files as reading material instead of instructions.
Reply: “Greet me and start the interview.”
If that fails, start a new chat, unzip the package, and upload every markdown file together.
Include the activation line from Step 1.
The interview never started
Confirm that visual-engine-core.md, visual-reference.md, and START-HERE.md were all loaded. A fresh conversation is usually faster than debugging a chat whose context began incorrectly.
The output looks polished but generic
Add two or three taste anchors and one explicit rejection. For example: “Use the restraint and negative space of these references. Avoid glossy AI drama, floating objects, neon rim light, and generic luxury styling.” Then regenerate once.
The product, food, person, or place is inaccurate
Stop generating the subject from scratch. Upload a real photo and ask for enhancement, relighting, environment replacement, or art direction while preserving the exact subject.
The poster text or price is wrong
Ask for a text-free substrate with reserved copy space. Add exact prices, dates, URLs, phone numbers, and legal text in Canva or another editor. A wrong number is not “nearly done.” It is wrong.
The image model hit a limit
Save the Brand Profile and the approved prompt. Resume in a new session or another supported image model. The Brand Profile preserves the system; the chat window is disposable.